Article
Remote Sensing
Ali Jamali, Masoud Mahdianpari, Fariba Mohammadimanesh, Saeid Homayouni
Summary: Wetlands are critical ecosystems that are in decline due to human activities and climate change. This study investigates the use of deep learning models and satellite data to efficiently map wetlands, and tests the proposed method in three study areas in Canada.
INTERNATIONAL JOURNAL OF APPLIED EARTH OBSERVATION AND GEOINFORMATION
(2022)
Article
Biotechnology & Applied Microbiology
Yongliang Chen, Chuan Lin, Yakun Qiao
Summary: In this study, we propose a two-pathway encoding network for edge detection. By combining Swin Transformer and deep separable convolution, our method outperforms the state-of-the-art CNN-based method on the BSDS500 dataset. Additionally, our proposed method has a computational advantage compared to the Transformer-based SOTA method in terms of FLOPs and FPS.
FRONTIERS IN BIOENGINEERING AND BIOTECHNOLOGY
(2022)
Article
Computer Science, Artificial Intelligence
Ludovica Schaerf, Eric Postma, Carina Popovici
Summary: In recent years, vision transformers, originally designed for language, have been successfully applied to visual tasks and shown to excel in various art authentication tasks. This paper compares the performance of Swin transformers and EfficientNet in art authentication using datasets of Vincent van Gogh's authentic paintings and contrast datasets. The results indicate that EfficientNet performs better in a standard contrast set, while Swin transformer achieves a higher authentication accuracy of over 85% in a contrast set consisting of only imitations. These findings suggest that vision transformers are promising contenders for enhancing computer-based art authentication, particularly in detecting artistic imitations.
NEURAL COMPUTING & APPLICATIONS
(2023)
Article
Computer Science, Information Systems
Shaheen Usmani, Sunil Kumar, Debanjan Sadhya
Summary: Deepfake is a deep learning-based technique for generating fake face images. This work proposes a shallow vision transformer model for detecting deepfakes, using an attention mechanism and a multi-head attention module to highlight important sections of deepfake images. The proposed model achieves high accuracy on deepfake detection datasets and outperforms existing state-of-the-art models. It is also applicable in scenarios with limited resources, making it valuable for practical use.
MULTIMEDIA TOOLS AND APPLICATIONS
(2023)
Article
Telecommunications
Lujie Cheng, Zhi Zhang, Chen Dong, Sirui Liu
Summary: In this letter, we propose a multi-slot conditional generative adversarial network (cGAN) called MSTGAN based on Swin Transformer for channel estimation in SISO scenario. The proposed MSTGAN learns the temporal correlation feature from continuous channel slots using 3D convolution and extracts deep features using Swin Transformer to improve the accuracy of channel estimation. The model is trained with data augmentation to reduce computational cost for offline training. Simulation results demonstrate that the proposed method outperforms LMMSE method and other deep learning methods. Furthermore, extension schemes of the MSTGAN to MIMO case are also provided.
IEEE COMMUNICATIONS LETTERS
(2023)
Article
Engineering, Electrical & Electronic
Jingzhi Tu, Gang Mei, Zhengjing Ma, Francesco Piccialli
Summary: In this study, the authors propose a high-resolution remote sensing image reconstruction method called SWCGAN, which combines convolutional and swin transformer layers to address the limitations of convolutional layers in modeling long-range dependencies. Experimental results show that the proposed method outperforms other state-of-the-art methods in terms of reconstruction performance.
IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING
(2022)
Article
Automation & Control Systems
Shanshan Huang, Xin Jin, Qian Jiang, Li Liu
Summary: This paper comprehensively reviews the latest progress in deep learning-based image colorization techniques, providing a systematic and comprehensive understanding of DLIC. The review covers algorithm classification, color space, network structure, loss function, level of automation, application fields, benchmark datasets, performance evaluation metrics, challenges, and future research directions. It serves as a valuable reference for researchers in image colorization and related fields.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2022)
Article
Remote Sensing
Ali Jamali, Swalpa Kumar Roy, Pedram Ghamisi
Summary: In this study, a deep learning algorithm called WetMapFormer is developed to accurately map wetlands using a combination of CNNs and Vision Transformers. By utilizing a local window attention mechanism, the algorithm improves feature generalization while reducing computational cost. Extensive evaluation in three pilot sites in Canada demonstrates the robustness of WetMapFormer.
INTERNATIONAL JOURNAL OF APPLIED EARTH OBSERVATION AND GEOINFORMATION
(2023)
Article
Computer Science, Artificial Intelligence
Ricard Lado-Roige, Marco A. Perez
Summary: The goal of video motion magnification techniques is to amplify small movements in videos that were previously invisible. This has applications in various fields such as biomedicine, deepfake detection, structural modal analysis, and predictive maintenance. However, distinguishing small motions from noise is challenging, especially when magnifying subtle, sub-pixel movements. This work introduces a state-of-the-art model based on the Swin Transformer that offers improved tolerance to noisy inputs and produces higher-quality outputs with less noise, blurriness, and artifacts compared to prior techniques. The improved output image quality enables more precise measurements for applications relying on magnified video sequences and may facilitate further advancements in video motion magnification techniques in new technical fields.
KNOWLEDGE-BASED SYSTEMS
(2023)
Article
Environmental Sciences
Ali Jamali, Masoud Mahdianpari
Summary: The emergence of deep learning techniques has revolutionized the classification of complicated environments, such as remote sensing. This study employed the Swin Transformer algorithm to classify complex coastal wetlands and compared its performance with the well-known deep CNNs of AlexNet and VGG-16. The results showed that the Swin Transformer algorithm outperformed the other techniques in terms of accuracy and F-1 scores, indicating the high capability of transformers in remote sensing for the classification of complex landscapes.
Article
Chemistry, Analytical
Bingcai Wei, Di Wang, Zhuang Wang, Liye Zhang
Summary: This study proposes a generative adversarial network with multiple attention mechanisms for image rain removal tasks. By integrating network structures and leveraging the powerful capabilities of deep learning, the method achieves successful completion of low-level visual tasks. The proposed approach outperforms state-of-the-art methods on synthetic and real image datasets.
Article
Computer Science, Interdisciplinary Applications
Kyeong-Beom Park, Jae Yeol Lee
Summary: This study proposes a novel deep learning model, SwinE-Net, for accurate colorectal polyp segmentation. The model effectively combines a CNN-based EfficientNet and Vision Transformer-based Swin Transformer. The proposed approach is evaluated and compared on multiple datasets, demonstrating its superior performance in polyp segmentation.
JOURNAL OF COMPUTATIONAL DESIGN AND ENGINEERING
(2022)
Article
Computer Science, Information Systems
Lintao Xu, Changhui Hu, Bo Zhang, Fei Wu, Ziyun Cai
Summary: This paper presents a method called STRN (Swin Transformer and ResNet-based Generative Adversarial Network) for enhancing low-light images. By combining the advantages of ResNet and the Swin Transformer, the proposed method achieves state-of-the-art performance in low-light image enhancement tasks according to both visual quality and evaluation metrics.
MULTIMEDIA TOOLS AND APPLICATIONS
(2023)
Article
Computer Science, Artificial Intelligence
Young-Jin Heo, Woon-Ha Yeo, Byung-Gyu Kim
Summary: This paper proposes an efficient vision transformer model for DeepFake detection that extracts both local and global features. Experimental results show that the proposed model outperforms existing models on multiple datasets.
APPLIED INTELLIGENCE
(2023)
Article
Computer Science, Information Systems
Chunhe Yu, Lingyue Hong, Tianpeng Pan, Yufeng Li, Tingting Li
Summary: The article proposes the ESTUGAN algorithm for remote sensing image super-resolution, which addresses the limitations of CNN-based algorithms and generative adversarial networks (GANs) in this task. By using Swin Transformer as the network backbone and employing a U-Net discriminator with region-aware learning strategy for assisted supervision, ESTUGAN achieves impressive performance in reconstructing remote sensing images.
Article
Remote Sensing
Ali Jamali, Masoud Mahdianpari, Brian Brisco, Jean Granger, Fariba Mohammadimanesh, Bahram Salehi
Summary: A timely and computationally efficient CNN architecture was proposed in this paper, which outperformed GoogleNet and SqueezeNet by about 12.71% and 12.2% in terms of mean overall accuracy in wetland classification.
CANADIAN JOURNAL OF REMOTE SENSING
(2021)
Article
Environmental Sciences
Ali Jamali, Masoud Mahdianpari
Summary: This study explores the potential and limitations of integrating deep learning networks with transformers for complex coastal wetland classification. The results demonstrate that the multi-model network outperforms other solo classifiers in wetland classification.
Article
Environmental Sciences
Ali Jamali, Masoud Mahdianpari
Summary: The emergence of deep learning techniques has revolutionized the classification of complicated environments, such as remote sensing. This study employed the Swin Transformer algorithm to classify complex coastal wetlands and compared its performance with the well-known deep CNNs of AlexNet and VGG-16. The results showed that the Swin Transformer algorithm outperformed the other techniques in terms of accuracy and F-1 scores, indicating the high capability of transformers in remote sensing for the classification of complex landscapes.
Article
Environmental Sciences
Ali Jamali, Masoud Mahdianpari, Fariba Mohammadimanesh, Brian Brisco, Bahram Salehi
Summary: Developed in response to the rapid loss or change of natural ecosystems, especially wetlands, due to human activities and climate change, this study presents a Deep Convolutional Neural Network (DCNN) model using a modified architecture of AlexNet and a Generative Adversarial Network (GAN) for wetland classification and generation of Sentinel-1 and Sentinel-2 data. Tested in a 370 sq. km area in Newfoundland, the proposed model achieved an average accuracy of 92.30% with improved F-1 scores for various wetland classes compared to the original CNN network of AlexNet. These results demonstrate the high capability of the proposed model for large-scale wetland classification tasks.
Article
Environmental Sciences
Ali Jamali
Summary: This research compared different optimizers for remote sensing image classification, with medium-sized neural network (MNN) showing the best performance, while Derivative-free Function Multi-layer Perceptron (DFMLP) performed best in 15m pixel-based Landsat-8 imagery.
EGYPTIAN JOURNAL OF REMOTE SENSING AND SPACE SCIENCES
(2021)
Article
Environmental Sciences
Mustafa Ibrahim Mohamed Elhaj, Tulay Ekemen Keskin, Ali Jamali
Summary: This study conducted meteorological drought analysis in the Great Man-Made River region in Libya, using the SPI and RDI methods to determine the severity of drought in different monitoring stations. The results showed that the years 2000-2001, 1981-1982, 1984-1985, and 1992-1993 experienced high drought rates in the region.
ENVIRONMENTAL EARTH SCIENCES
(2022)
Article
Geochemistry & Geophysics
Ali Jamali, Swalpa Kumar Roy, Avik Bhattacharya, Pedram Ghamisi
Summary: Convolutional neural networks (CNNs) have achieved great success in image classification, but the use of transformers in remote sensing, especially with limited labeled data, remains challenging. In this research, we propose a vision transformer (ViT)-based framework that combines 3D and 2D CNNs as feature extractors and utilizes local window attention (LWA) for effective classification of PolSAR data. Experimental results show that our model, PolSARFormer, outperforms state-of-the-art algorithms like Swin Transformer and FNet, achieving higher classification accuracy in both the San Francisco and Flevoland datasets.
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS
(2023)
Article
Remote Sensing
Ali Jamali, Swalpa Kumar Roy, Jonathan Li, Pedram Ghamisi
Summary: The research introduces a TransU-Net++ model aided by attention gates for semantic segmentation of deforestation in South American forest biomes. The TransU-Net++ improves the performance of TransU-Net by about 4%, 6%, and 16% in terms of overall accuracy, F1-score, and recall, respectively, on the Atlantic Forest dataset. Furthermore, it achieves the highest Area under the ROC Curve value compared to other segmentation models on the 3-band Amazon forest dataset.
INTERNATIONAL JOURNAL OF APPLIED EARTH OBSERVATION AND GEOINFORMATION
(2023)
Article
Remote Sensing
Ali Jamali, Swalpa Kumar Roy, Pedram Ghamisi
Summary: In this study, a deep learning algorithm called WetMapFormer is developed to accurately map wetlands using a combination of CNNs and Vision Transformers. By utilizing a local window attention mechanism, the algorithm improves feature generalization while reducing computational cost. Extensive evaluation in three pilot sites in Canada demonstrates the robustness of WetMapFormer.
INTERNATIONAL JOURNAL OF APPLIED EARTH OBSERVATION AND GEOINFORMATION
(2023)
Proceedings Paper
Geosciences, Multidisciplinary
Ali Jamali, Fariba Mohammadimanesh, Masoud Mahdianpari
Summary: This study demonstrates the effectiveness of the advanced Swin Transformer for complex wetland classification, achieving high recognition accuracy for various wetland types.
2022 IEEE INTERNATIONAL GEOSCIENCE AND REMOTE SENSING SYMPOSIUM (IGARSS 2022)
(2022)
Article
Remote Sensing
Ali Jamali, Masoud Mahdianpari, Fariba Mohammadimanesh, Saeid Homayouni
Summary: Wetlands are critical ecosystems that are in decline due to human activities and climate change. This study investigates the use of deep learning models and satellite data to efficiently map wetlands, and tests the proposed method in three study areas in Canada.
INTERNATIONAL JOURNAL OF APPLIED EARTH OBSERVATION AND GEOINFORMATION
(2022)
Proceedings Paper
Computer Science, Interdisciplinary Applications
Mustafa Ibrahim Mohamed Elhaj, Tulay Ekemen Keskin, Ali Jamali
Summary: This study conducted meteorological drought analysis for five monitoring stations in Northern Libya using the SPI and RDI methods. The driest and least dry periods in the region were determined, as well as the longest drought period.
CLIMATE CHANGE, NATURAL RESOURCES AND SUSTAINABLE ENVIRONMENTAL MANAGEMENT
(2022)
Article
Engineering, Electrical & Electronic
Ali Jamali, Masoud Mahdianpari, Fariba Mohammadimanesh, Brian Brisco, Bahram Salehi
Summary: This article introduces a novel approach called 3-D hybrid GAN to address the limited training sample issue in classification, achieving better results in complex wetland classification.
IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING
(2022)
Article
Geochemistry & Geophysics
Ali Jamali, Masoud Mahdianpari, Fariba Mohammadimanesh, Avik Bhattacharya, Saeid Homayouni
Summary: This article proposes the use of Haar wavelet transform in deep CNNs for improved classification accuracy of PolSAR imagery. Experimental results show that the proposed method outperforms other shallow CNN models in terms of accuracy and consistency of classification results.
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS
(2022)
